CN115586506B - Anti-interference target classification method and device - Google Patents
Anti-interference target classification method and device Download PDFInfo
- Publication number
- CN115586506B CN115586506B CN202211594912.7A CN202211594912A CN115586506B CN 115586506 B CN115586506 B CN 115586506B CN 202211594912 A CN202211594912 A CN 202211594912A CN 115586506 B CN115586506 B CN 115586506B
- Authority
- CN
- China
- Prior art keywords
- point cloud
- neural network
- cloud information
- classification
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/414—Discriminating targets with respect to background clutter
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/417—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section involving the use of neural networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The application provides an anti-interference target classification method and device, which comprise the following steps: acquiring a plurality of millimeter wave radar point cloud information; inputting the point cloud information of the millimeter wave radar into a target classification neural network to obtain a plurality of classification results; and obtaining a target classification result according to the plurality of classification results. Therefore, the millimeter wave radar is slightly influenced by the natural environment, so that the millimeter wave radar point cloud is utilized for target classification, the target detection time can be shortened, the working performance of target classification under complex conditions is improved, and the millimeter wave radar point cloud has certain anti-interference capability on the complex environment. Meanwhile, classification results of a plurality of millimeter wave radar point clouds are combined, so that the problem that the classification result is inaccurate due to interference of a single millimeter wave radar point cloud can be solved.
Description
Technical Field
The application relates to the technical field of artificial intelligence, in particular to an anti-interference target classification method and device.
Background
The automatic driving automobile is cooperated with artificial intelligence, visual calculation, radar, monitoring device, global positioning system and other technology to make the computer operate the motor vehicle automatically and safely without any active operation of human. In the automatic driving process, the most important is to realize the active safety anti-collision function, such as the functions of front vehicle anti-collision early warning, lane change assistance, adaptive cruise control, blind spot monitoring and the like.
The conventional automatic driving function requires a camera to assist in the identification of the kind of the object, for example, which kind of car, pedestrian, bicycle, etc. the object is. Because under heavy rain, heavy fog weather, camera system receives the interference easily, leads to its unable normal work, so, utilizes the supplementary affirmation of carrying out the target kind of camera, influences the performance of initiative safe anticollision very easily.
Therefore, how to provide a target classification method which is not easily interfered is a technical problem which needs to be solved urgently by those skilled in the art.
Disclosure of Invention
In view of this, embodiments of the present application provide an anti-interference target classification method and apparatus, and aim to provide a target classification method with little influence from the environment.
In a first aspect, an embodiment of the present application provides an anti-interference target classification method, including:
acquiring a plurality of millimeter wave radar point cloud information;
inputting the millimeter wave radar point cloud information into a target classification neural network to obtain a plurality of classification results;
and obtaining a target classification result according to the plurality of classification results.
Optionally, before the obtaining the multiple pieces of millimeter wave radar point cloud information, the method further includes:
constructing a pre-training neural network, wherein the pre-training neural network comprises a convolutional layer, a maximum pooling layer and a full-connection layer;
and training the pre-training neural network by using a training sample to obtain the target classification neural network, wherein the target classification neural network comprises an iteration convolution layer, an iteration maximum pooling layer and an iteration full-connection layer.
Optionally, the training samples include sample information and result information;
the training of the pre-training neural network by using the training samples to obtain the target classification neural network comprises:
inputting the sample information into the pre-training neural network to obtain a training result;
obtaining the convergence degree of the pre-training neural network according to the training result and the result information;
in response to that the convergence degree of the pre-training neural network does not reach a preset condition, correcting the weight of the pre-training neural network by using a gradient descent method to obtain an iterative neural network;
and replacing the pre-training neural network with the iterative neural network, returning and executing the input of the sample information into the pre-training neural network to obtain a training result until the convergence degree of the pre-training neural network reaches a preset condition, and determining the iterative neural network as the target classification neural network.
Optionally, the step of inputting the point cloud information of the millimeter wave radar into a target classification neural network to obtain a plurality of classification results includes:
inputting the millimeter wave radar point cloud information into the iterative convolution layer to obtain a plurality of ascending-dimensional point cloud information;
inputting the plurality of ascending-dimensional point cloud information into the iteration maximum pooling layer to obtain a plurality of pooled point cloud information;
and inputting the plurality of pooled point cloud information into the iteration full-link layer to obtain a plurality of classification results.
Optionally, the inputting the multiple millimeter wave radar point cloud information into the iterative convolution layer to obtain multiple ascending-dimensional point cloud information includes:
inputting the millimeter wave radar point cloud information into a first convolution layer to obtain a plurality of basic ascending dimensional point cloud information;
and inputting the millimeter wave radar point cloud information into a second convolution layer to obtain a plurality of ascending-dimensional point cloud information.
Optionally, after the millimeter wave radar point cloud information is input into the first convolution layer to obtain a plurality of basic ascending-dimension point cloud information, the method further includes:
splicing the basic ascending-dimensional point cloud information to obtain spliced point cloud information;
inputting the plurality of spliced point cloud information into a second convolution layer to obtain a plurality of spliced ascending-dimensional point cloud information;
inputting the multiple spliced dimensionality-increased point cloud information into the iteration maximum pooling layer to obtain multiple spliced pooled point cloud information;
inputting the plurality of spliced pooling point cloud information into the iteration full-connection layer to obtain a plurality of spliced classification results;
after obtaining a target classification result according to the plurality of classification results, the method further includes:
and obtaining a comprehensive target classification result according to the plurality of classification results and the plurality of splicing classification results.
Optionally, the obtaining a target classification result according to the plurality of classification results includes:
in response to the plurality of classification results being consistent, confirming that the target classification result is any one of the plurality of classification results;
and in response to the plurality of classification results being inconsistent, confirming that the target classification result is not credible.
In a second aspect, an embodiment of the present application provides an anti-interference target classification apparatus, including:
the acquisition module is used for acquiring a plurality of millimeter wave radar point cloud information;
the first classification module is used for inputting the point cloud information of the millimeter wave radar into a target classification neural network to obtain a plurality of classification results;
and the second classification module is used for obtaining a target classification result according to the plurality of classification results.
In a third aspect, an embodiment of the present application provides an apparatus, where the apparatus includes a memory and a processor, where the memory is configured to store instructions or codes, and the processor is configured to execute the instructions or codes, so as to cause the apparatus to perform the interference rejection target classification method according to any one of the foregoing first aspects.
In a fourth aspect, an embodiment of the present application provides a computer storage medium, where codes are stored in the computer storage medium, and when the codes are executed, an apparatus that executes the codes implements the interference-free target classification method according to any one of the foregoing first aspects.
The embodiment of the application provides an anti-interference target classification method and device, and when the method is executed, a plurality of millimeter wave radar point cloud information are obtained firstly; then inputting the point cloud information of the millimeter wave radar into a target classification neural network to obtain a plurality of classification results; and finally, obtaining a target classification result according to the plurality of classification results. Therefore, the influence of the natural environment on the millimeter wave radar is small, the millimeter wave radar point cloud is utilized for target classification, the target detection time can be shortened, the working performance of target classification under complex conditions is improved, and the millimeter wave radar point cloud has certain anti-interference capability on the complex environment. Meanwhile, classification results of a plurality of millimeter wave radar point clouds are combined, so that the problem that the classification result is inaccurate due to interference of a single millimeter wave radar point cloud can be solved.
Drawings
To illustrate the technical solutions in the present embodiment or the prior art more clearly, the drawings needed to be used in the description of the embodiment or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a flowchart of a method of an anti-interference target classification method according to an embodiment of the present application;
fig. 2 is a schematic diagram of point cloud information of a plurality of millimeter wave radars according to an embodiment of the present disclosure;
fig. 3 is a flowchart of another method of an anti-interference target classification method according to an embodiment of the present application;
fig. 4 is a schematic diagram of a neural network of an anti-interference target classification method according to an embodiment of the present application;
fig. 5 is a flowchart of another method of an anti-interference target classification method according to an embodiment of the present application;
FIG. 6 is a schematic diagram of another neural network of the method for classifying an anti-interference target according to the present embodiment;
fig. 7 is a schematic diagram of feature concatenation of an anti-interference target classification method according to an embodiment of the present application;
fig. 8 is a schematic structural diagram of an anti-interference object classification apparatus according to an embodiment of the present application.
Detailed Description
In the automatic driving process, the most important is to realize the active safety anti-collision function, for example, the functions of front vehicle anti-collision early warning, lane change assistance, adaptive cruise control, blind spot monitoring and the like. In the conventional automatic driving function, the camera is used to assist the identification of the target category, for example, the identification of the target object in which category the target object is, such as cars, pedestrians, bicycles, etc. Because the camera system can not work normally in heavy rain and heavy fog weather, the camera is used for assisting in confirming the target type, and the performance of active safety collision avoidance is easily influenced.
The method provided by the embodiment of the application is executed by computer equipment and is used for providing a target classification method which is less influenced by environment.
It should be apparent that the described embodiments are only a few embodiments of the present application, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
Referring to fig. 1, fig. 1 is a flowchart of a method of an anti-interference target classification method according to an embodiment of the present application, where the method includes:
step S101: and acquiring a plurality of millimeter wave radar point cloud information.
The millimeter wave radar generally transmits frequency modulation continuous waves in each pulse period, and mixes the echoes at a receiving end, and the final obtained result is superposition of a plurality of target echoes. The superimposed one-dimensional echo signal is fourier-transformed to obtain a one-dimensional echo image, also referred to as a one-dimensional range profile. Further, fourier transform is performed on a one-dimensional range profile obtained by a plurality of continuously transmitted echoes in another dimension, and finally distribution of a target in an observation space on a range-doppler velocity image can be obtained, as shown in fig. 2, fig. 2 is a schematic view of a plurality of millimeter wave radar point cloud information of the anti-interference target classification method provided in the embodiment of the present application. Each target point appears as a bright spot on the graph, the position of the bright spot reflects the distance and the speed of the target, and the intensity reflects the echo intensity of the target. The multiple target points and the information represented by the target points jointly form single millimeter wave radar point cloud information.
Existing target detection algorithms are based on cameras or lidar, but both are susceptible to severe weather or lighting conditions. The wavelength of the millimeter wave radar is generally between 1-19mm, which is between centimeter wave and light wave, and has the advantages of microwave guidance and photoelectric guidance, and the guide head has the characteristics of small volume, light weight and high spatial resolution, and has strong capability of penetrating fog, smoke and dust, and is not easily influenced by severe weather or illumination conditions. But because of its lower precision, it needs to be further processed to achieve better classification effect.
Step S102: and inputting the point cloud information of the millimeter wave radar into a target classification neural network to obtain a plurality of classification results.
The target classification neural network can comprise operations such as convolution, dimensionality raising, full connection and the like, and is used for extracting the characteristics of the millimeter wave radar information and analyzing the extracted characteristics to obtain the classification of the millimeter wave radar. As a possible implementation mode, the target classification neural network can be built through the pyroch, and then the target classification neural network is tested and iterated to make the target classification neural network converged, so that the classification accuracy is improved.
Because the resolution and precision of the point cloud information of a single millimeter wave radar may be low, the point cloud information of a plurality of millimeter wave radars needs to be input into the target classification neural network to obtain a plurality of classification results. And analyzing and comparing the classification result to obtain a credible classification result.
Step S103: and obtaining a target classification result according to the plurality of classification results.
The millimeter wave radar point cloud resolution is low, so that the classification result is likely to be changed into an interfered result, and different classification results are different for the same target. Therefore, a plurality of classification results need to be compared to obtain the same classification result, and the classification result is adopted as a credible classification result, so as to achieve the purpose of anti-interference of the classification result.
In summary, in this embodiment, because the influence of the natural environment on the millimeter wave radar is small, the millimeter wave radar point cloud is used for target classification, so that the target detection time can be shortened, the working performance of target classification under complex conditions is improved, and the millimeter wave radar point cloud has a certain anti-interference capability to the complex environment. Meanwhile, classification results of a plurality of millimeter wave radar point clouds are combined, so that the problem that the classification result is inaccurate due to interference of a single millimeter wave radar point cloud can be solved.
In the embodiment of the present application, there are many possible implementations of the steps described in fig. 1, which are described below separately. It should be noted that the implementation manners given in the following description are only exemplary illustrations, and do not represent all implementation manners of the embodiments of the present application.
Referring to fig. 3, this figure is a flowchart of another method of the interference-free target classification method according to the embodiment of the present application, including:
step S301: and constructing a pre-training neural network.
The pre-training neural network can comprise a convolutional layer, a maximum pooling layer, a full-link layer and the like and is used for classifying the point cloud information after feature extraction. Referring to fig. 4, which is a schematic diagram of a neural network of the method for classifying an anti-interference target according to the embodiment of the present application, three layers of processing on the right side of the input two-dimensional data in the diagram respectively represent a convolutional layer h, a max-pooling layer g, and a full-link layer γ.
Step S302: and training the pre-training neural network by using a training sample to obtain the target classification neural network.
The training samples are samples that include sample information and result information. As a possible embodiment, the sample information may be measured data of which the data set is nuscience public data set, and the result information may be information for labeling the target with a standard box. As another possible implementation, the training sample may also be millimeter wave radar point cloud data obtained after autocad modeling and simulation.
Specifically, the process of training the pre-trained neural network by using the training samples to obtain the target classification neural network includes:
step S3021: and inputting the sample information into the pre-training neural network to obtain a training result.
The pre-trained neural network can complete classified work, but the generalization capability is poor, sample information needs to be input into the pre-trained neural network to obtain a training result, and then the pre-trained neural network is adjusted based on the training result.
Step S3022: and obtaining the convergence degree of the pre-training neural network according to the training result and the result information.
And comparing the training result with the result information, and evaluating the output value of the neural network by using the actual value so as to obtain the convergence degree of the pre-training neural network.
Step S3023: and in response to the fact that the convergence degree of the pre-training neural network does not reach a preset condition, correcting the weight of the pre-training neural network by using a gradient descent method to obtain an iterative neural network.
When the convergence degree of the pre-training neural network does not reach the preset condition, it is indicated that the precision of the pre-training neural network is low, the obtained result is not accurate, and the related weight needs to be adjusted to improve the precision. As a possible implementation, the weights of the pre-trained neural network may be modified by a gradient descent method.
The adjusted pre-training neural network is an iterative neural network, namely the adjusted convolutional layer is an iterative convolutional layer, the adjusted maximum pooling layer is an iterative maximum pooling layer, and the adjusted full-link layer is an iterative full-link layer.
Step S3024: and replacing the pre-trained neural network with the iterative neural network, returning to execute the step of inputting the sample information into the pre-trained neural network to obtain a training result until the convergence degree of the pre-trained neural network reaches a preset condition, and determining the iterative neural network as the target classification neural network.
And replacing the pre-training neural network with the iterative neural network, and repeatedly executing the steps S3021-S3022, so that the neural network can be iterated for multiple times, and the convergence degree and accuracy of the neural network can be gradually improved.
And when the convergence degree of the pre-trained neural network reaches a preset condition, the output accuracy of the pre-trained neural network is higher, the pre-trained neural network is more fit to reality and can be put into use, and the result is credible, and at the moment, the iterative neural network is determined to be the target classification neural network for subsequent use.
Step S303: and acquiring a plurality of millimeter wave radar point cloud information.
Step S304: and inputting the millimeter wave radar point cloud information into the iterative convolution layer to obtain a plurality of ascending-dimensional point cloud information.
And inputting the millimeter wave radar point cloud information into the iterative convolution layer, and performing feature extraction on the millimeter wave radar point cloud information by using the iterative convolution layer.
As a possible implementation mode, when a plurality of millimeter wave radar point cloud information input is represented as (N, C) in L), representing the output multiple ascending-dimensional point cloud information as (N, C) out ，L out ) The feature extraction may be formulated as:
wherein N is the number of millimeter wave radar point cloud information or a plurality of ascending-dimensional point cloud information, C out The number of channels of a plurality of ascending-dimensional point cloud information, bias is the offset of the iterative convolution layer, C in The channel number of the millimeter wave radar point cloud information is multiple, the weight is the weight of the iterative convolution layer, and the input is the input convolution shape.
Wherein L is out The data length of the point cloud information of multiple ascending dimensions, L the data length of the point cloud information of multiple millimeter wave radars, padding the filling size of the point cloud information of multiple millimeter wave radars, partition the size between convolution kernels, kernel _ size the size of the convolution kernels, and stride the moving step size of the convolution kernels.
Step S305: and inputting the plurality of ascending-dimensional point cloud information into the iteration maximum pooling layer to obtain a plurality of pooled point cloud information.
The iterative maximum pooling layer may be referred to a pooling network commonly used in the art, and is not limited herein.
Step S306: and inputting the plurality of pooled point cloud information into the iteration full-link layer to obtain a plurality of classification results.
The iterative fully-connected layer may be referred to a fully-connected network commonly used in the art, and is not limited herein.
As a possible implementation, steps S304 to S306 may be represented by combining:
wherein f represents the calculation of the millimeter wave radar point cloud information by the target classification neural network, and x 1 ，x 2 ，x n ，
，
，
The point cloud information of the single millimeter wave radar is respectively, gamma represents the calculation of an iteration full-link layer, g represents the calculation of an iteration maximum pooling layer, and h represents the calculation of an iteration convolution layer.
It should be noted that, because the millimeter wave radar point cloud information has a disorder property, the millimeter wave radar point cloud information is input into the iterative convolution layer without being affected by the order, and can be expressed as:
wherein x is 1 ，x 2 ，x n ，
，
，
The point cloud information of the millimeter wave radar is single, and f represents the calculation of the point cloud information of the millimeter wave radar by the target classification neural network.
Step S307: and obtaining a target classification result according to the plurality of classification results.
The millimeter wave radar point cloud resolution is low, so that the classification result is likely to become an interfered result, and different classification results are different for the same target. Therefore, a plurality of classification results need to be compared to obtain the same classification result, and the classification result is adopted as a credible classification result, so as to achieve the purpose of anti-interference of the classification result.
Therefore, in response to the plurality of classification results being consistent, confirming the target classification result as any one of the plurality of classification results; and in response to the plurality of classification results being inconsistent, confirming that the target classification result is not credible.
In summary, in the embodiment, the target classification neural network with a good classification effect can be obtained by training the target classification neural network, so that the accuracy and the classification efficiency of the classification result are improved; the convolutional network is refined, the convolutional network with a good feature extraction effect is provided, the working performance of target classification under complex conditions can be improved, the classification accuracy is greatly improved, and the convolutional network has certain anti-interference capability.
In the embodiment of the present application, there are many possible implementations of the steps described in fig. 1, which are described below separately. It should be noted that the implementation manners given in the following description are only exemplary illustrations, and do not represent all implementation manners of the embodiments of the present application.
Referring to fig. 5, this figure is a flowchart of another method of the interference-free target classification method according to the embodiment of the present application, including:
step S501: and constructing a pre-training neural network.
Step S502: and training the pre-training neural network by using a training sample to obtain the target classification neural network.
Step S503: and acquiring a plurality of millimeter wave radar point cloud information.
Step S504: and inputting the millimeter wave radar point cloud information into a first convolution layer to obtain a plurality of basic ascending dimensional point cloud information.
As a possible implementation manner, the target classification neural network may be as shown in fig. 6, where fig. 6 is another schematic diagram of the neural network of the anti-interference target classification method provided in the embodiment of the present application.
The iterative convolution layer can comprise a plurality of convolution layers, in order to combine different millimeter wave radar point cloud information and acquire the relation between points, the millimeter wave radar point cloud information can be first raised into a plurality of basic raised-dimension point cloud information through the first convolution layer, so that the millimeter wave radar point cloud information can be spliced together subsequently.
Step S505: and splicing the basic ascending-dimensional point cloud information to obtain a plurality of spliced point cloud information.
The basic ascending-dimensional point cloud information is spliced together to obtain a plurality of spliced point cloud information, the spliced point cloud information is subjected to subsequent operation, a classification result considering the relation between points can be obtained, and the anti-interference performance of the classification result is further improved. For example, the convolved features n x and n y are subjected to stitching expansion to form n x (x + y) features.
As a possible implementation manner, during the stitching, the point clouds in different radii can be searched and then stitched. For example, referring to fig. 7, fig. 7 is a schematic diagram of feature concatenation of the anti-interference object classification method according to the embodiment of the present application. For example, the picture is divided into different rings by three circles with different radiuses, each ring is internally provided with a plurality of point clouds, and basic ascending-dimensional point cloud information of the point clouds in the different rings is spliced together, so that the anti-interference function can be realized.
As another possible implementation, different splicing schemes can be divided by a kd-tree algorithm.
Step S506: classifying the plurality of stitched point cloud information based on a neural network.
Firstly, inputting the plurality of spliced point cloud information into a second convolution layer to obtain a plurality of spliced dimensionality-increased point cloud information; inputting the spliced ascending dimensional point cloud information into the iteration maximum pooling layer to obtain spliced pooled point cloud information; and finally, inputting the information of the plurality of spliced pooling point clouds into the iteration full-connection layer to obtain a plurality of splicing classification results.
The above contents are the same as the above embodiments, and are not described again here.
Step S507: classifying the plurality of millimeter wave radar point cloud information based on a neural network.
Firstly, inputting the millimeter wave radar point cloud information into a second convolution layer to obtain a plurality of ascending-dimensional point cloud information; inputting the plurality of ascending-dimension point cloud information into the iteration maximum pooling layer to obtain a plurality of pooled point cloud information; and finally, inputting the information of the plurality of pooled point clouds into the iteration full-link layer to obtain a plurality of classification results.
The above contents are the same as the above embodiments, and are not described again here.
Step S508: and obtaining a comprehensive target classification result according to the plurality of classification results and the plurality of splicing classification results.
The millimeter wave radar point cloud resolution is low, so that the classification result is likely to become an interfered result, and different classification results are different for the same target. In addition, only the millimeter wave radar point clouds are considered, and the relation between the point clouds cannot be considered.
Therefore, the plurality of classification results and the plurality of spliced classification results are compared, when the results are completely the same, the classification results are not interfered, the classification results are credible, otherwise, the classification results are considered to be credible. In response to the plurality of classification results and the plurality of splicing classification results being consistent, confirming that the comprehensive target classification result is any one of the plurality of classification results and the plurality of splicing classification results; and in response to the plurality of classification results and the plurality of spliced classification results being inconsistent, determining that the comprehensive target classification result is unreliable.
In summary, in this embodiment, the features after convolution are spliced, and a more accurate target classification result can be obtained by considering the relation between different point clouds, so that the anti-interference capability of the classification model is fully improved.
The foregoing provides some specific implementation manners of the anti-interference target classification method for the embodiments of the present application, and based on this, the present application also provides a corresponding apparatus. The device provided by the embodiment of the present application will be described in terms of functional modularity.
Referring to fig. 8, a schematic structural diagram of an apparatus for classifying an interference-free object includes an obtaining module 801, a first classifying module 802, and a second classifying module 803.
An obtaining module 801, configured to obtain point cloud information of multiple millimeter-wave radar points;
the first classification module 802 is configured to input the millimeter wave radar point cloud information into a target classification neural network to obtain a plurality of classification results;
and a second classification module 803, configured to obtain a target classification result according to the multiple classification results.
As a possible implementation, the apparatus further comprises:
the network construction module is used for constructing a pre-training neural network, and the pre-training neural network comprises a convolutional layer, a maximum pooling layer and a full-connection layer;
and the training module is used for training the pre-training neural network by using a training sample to obtain the target classification neural network, and the target classification neural network comprises an iteration convolution layer, an iteration maximum pooling layer and an iteration full-connection layer.
As a possible implementation, the training sample includes sample information and result information;
the training module comprises:
the training unit is used for inputting the sample information into the pre-training neural network to obtain training results;
a convergence judging unit, configured to obtain a convergence degree of the pre-training neural network according to the training result and the result information;
the correcting unit is used for responding that the convergence degree of the pre-training neural network does not reach a preset condition, and correcting the weight of the pre-training neural network by using a gradient descent method to obtain an iterative neural network;
and the iteration unit is used for replacing the pre-training neural network with the iterative neural network, returning and executing the input of the sample information into the pre-training neural network to obtain a training result until the convergence degree of the pre-training neural network reaches a preset condition, and determining the iterative neural network as the target classification neural network.
As a possible implementation, the first classification module 302 includes:
the dimension increasing unit is used for inputting the millimeter wave radar point cloud information into the iterative convolution layer to obtain a plurality of dimension increasing point cloud information;
the pooling unit is used for inputting the plurality of ascending-dimensional point cloud information into the iteration maximum pooling layer to obtain a plurality of pooled point cloud information;
and the full connection unit is used for inputting the plurality of pooled point cloud information into the iteration full connection layer to obtain a plurality of classification results.
As a possible implementation, the dimension-raising unit includes:
the first dimensionality increasing component is used for inputting the millimeter wave radar point cloud information into a first convolution layer to obtain a plurality of basic dimensionality increasing point cloud information;
and the second dimension-increasing component is used for inputting the millimeter wave radar point cloud information into a second convolution layer to obtain a plurality of dimension-increasing point cloud information.
As a possible implementation, the apparatus further comprises:
the splicing component is used for splicing the plurality of ascending-dimensional point cloud information to obtain a plurality of spliced point cloud information;
the splicing dimensionality increasing component is used for inputting the plurality of splicing point cloud information into a second convolution layer to obtain a plurality of splicing dimensionality increasing point cloud information;
the splicing pooling component is used for inputting the splicing ascending-dimensional point cloud information into the iteration maximum pooling layer to obtain splicing pooling point cloud information;
the splicing full-connection assembly is used for inputting the plurality of splicing pooled point cloud information into the iteration full-connection layer to obtain a plurality of splicing classification results;
and the comprehensive classification result component is used for obtaining a comprehensive target classification result according to the plurality of classification results and the plurality of splicing classification results.
As a possible implementation, the second classification module 303 includes:
a first confirming unit, configured to, in response to a correspondence between the plurality of classification results, confirm that the target classification result is any one of the plurality of classification results;
and the second confirming unit is used for responding to the inconsistency of the plurality of classification results and confirming that the target classification result is an unreliable result.
The embodiment of the application also provides corresponding equipment and a computer storage medium, which are used for realizing the scheme provided by the embodiment of the application.
The device comprises a memory and a processor, wherein the memory is used for storing instructions or codes, and the processor is used for executing the instructions or codes so as to enable the device to execute the anti-interference target classification method in any embodiment of the application.
The computer storage medium has code stored therein, and when the code is executed, a device running the code implements the method for classifying an anti-interference object according to any embodiment of the present application.
In the embodiments of the present application, the names "first" and "second" (if present) in the names "first" and "second" are used for name identification, and do not represent the first and second in sequence.
As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that all or part of the steps in the method of the above embodiments may be implemented by software plus a general hardware platform. Based on such understanding, the technical solution of the present application may be embodied in the form of a software product, which may be stored in a storage medium, such as a read-only memory (ROM)/RAM, a magnetic disk, an optical disk, or the like, and includes several instructions for enabling a computer device (which may be a personal computer, a server, or a network communication device such as a router) to execute the method according to the embodiments or some parts of the embodiments of the present application.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the apparatus embodiment, since it is substantially similar to the method embodiment, it is relatively simple to describe, and reference may be made to some descriptions of the method embodiment for relevant points. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
The above description is only an exemplary embodiment of the present application, and is not intended to limit the scope of the present application.
Claims (8)
1. An anti-interference object classification method, the method comprising:
acquiring point cloud information of a plurality of millimeter wave radars;
inputting the millimeter wave radar point cloud information into a first convolution layer to obtain a plurality of basic ascending dimensional point cloud information;
splicing the basic ascending-dimensional point cloud information to obtain spliced point cloud information;
inputting the millimeter wave radar point cloud information and the spliced point cloud information into a target classification neural network to obtain a plurality of classification results and a plurality of spliced classification results;
obtaining a target classification result according to the plurality of classification results;
and obtaining a comprehensive target classification result according to the plurality of classification results and the plurality of splicing classification results.
2. The method of claim 1, wherein prior to the obtaining the plurality of millimeter wave radar point cloud information, the method further comprises:
constructing a pre-training neural network, wherein the pre-training neural network comprises a convolutional layer, a maximum pooling layer and a full-connection layer;
training the pre-training neural network by using a training sample to obtain the target classification neural network, wherein the target classification neural network comprises an iteration convolutional layer, an iteration maximum pooling layer and an iteration full-connection layer, and the iteration convolutional layer comprises the first convolutional layer and a second convolutional layer.
3. The method of claim 2, wherein the training samples comprise sample information and result information;
the training the pre-training neural network by using the training sample to obtain the target classification neural network comprises:
inputting the sample information into the pre-training neural network to obtain a training result;
obtaining the convergence degree of the pre-training neural network according to the training result and the result information;
in response to that the convergence degree of the pre-training neural network does not reach a preset condition, correcting the weight of the pre-training neural network by using a gradient descent method to obtain an iterative neural network;
and replacing the pre-training neural network with the iterative neural network, returning and executing the input of the sample information into the pre-training neural network to obtain a training result until the convergence degree of the pre-training neural network reaches a preset condition, and determining the iterative neural network as the target classification neural network.
4. The method of claim 2, wherein the inputting the millimeter wave radar point cloud information and the stitched point cloud information into a target classification neural network to obtain a plurality of classification results and a plurality of stitched classification results comprises:
inputting the millimeter wave radar point cloud information into the second convolution layer to obtain a plurality of ascending-dimensional point cloud information;
inputting the plurality of spliced point cloud information into the second convolution layer to obtain a plurality of spliced ascending-dimension point cloud information;
inputting the plurality of ascending-dimensional point cloud information into the iteration maximum pooling layer to obtain a plurality of pooled point cloud information;
inputting the multiple spliced dimensionality-increased point cloud information into the iteration maximum pooling layer to obtain multiple spliced pooled point cloud information;
inputting the plurality of pooling point cloud information into the iteration full-connection layer to obtain a plurality of classification results;
and inputting the spliced pooling point cloud information into the iteration full-connection layer to obtain a plurality of spliced classification results.
5. The method of claim 1, wherein obtaining a target classification result from the plurality of classification results comprises:
in response to the plurality of classification results being consistent, confirming that the target classification result is any one of the plurality of classification results;
and in response to the plurality of classification results being inconsistent, confirming that the target classification result is not credible.
6. An apparatus for classifying an interference-resistant object, the apparatus comprising:
the acquisition module is used for acquiring a plurality of millimeter wave radar point cloud information;
the first dimensionality increasing component is used for inputting the millimeter wave radar point cloud information into a first convolution layer to obtain a plurality of basic dimensionality increasing point cloud information;
the splicing component is used for splicing the plurality of ascending-dimensional point cloud information to obtain a plurality of spliced point cloud information;
the classification module is used for inputting the millimeter wave radar point cloud information and the spliced point cloud information into a target classification neural network to obtain a plurality of classification results and spliced classification results;
the second classification module is used for obtaining a target classification result according to the plurality of classification results;
and the comprehensive classification result component is used for obtaining a comprehensive target classification result according to the plurality of classification results and the plurality of splicing classification results.
7. An apparatus comprising a memory for storing instructions or code and a processor for executing the instructions or code to cause the apparatus to perform the tamper resistant object classification method of any one of claims 1 to 5.
8. A computer storage medium having code stored therein, wherein when the code is executed, a computer storage device executing the code implements the tamper-resistant object classification method of any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211594912.7A CN115586506B (en) | 2022-12-13 | 2022-12-13 | Anti-interference target classification method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211594912.7A CN115586506B (en) | 2022-12-13 | 2022-12-13 | Anti-interference target classification method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115586506A CN115586506A (en) | 2023-01-10 |
| CN115586506B true CN115586506B (en) | 2023-03-17 |
Family
ID=84782961
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211594912.7A Active CN115586506B (en) | 2022-12-13 | 2022-12-13 | Anti-interference target classification method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115586506B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117111013B (en) * | 2023-08-22 | 2024-04-30 | 南京慧尔视智能科技有限公司 | Radar target tracking track starting method, device, equipment and medium |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107633220A (en) * | 2017-09-13 | 2018-01-26 | 吉林大学 | A kind of vehicle front target identification method based on convolutional neural networks |
| CN109086679A (en) * | 2018-07-10 | 2018-12-25 | 西安恒帆电子科技有限公司 | A kind of millimetre-wave radar safety check instrument foreign matter detecting method |
| CN110826450A (en) * | 2019-10-30 | 2020-02-21 | 北京无线电计量测试研究所 | Automatic suspicious article detection method based on millimeter wave image |
| CN112348056A (en) * | 2020-10-16 | 2021-02-09 | 北京大学深圳研究生院 | Point cloud data classification method, device, equipment and readable storage medium |
| CN114994665A (en) * | 2021-03-01 | 2022-09-02 | 武汉智行者科技有限公司 | Millimeter wave radar point cloud classification method |
| CN113516052B (en) * | 2021-05-21 | 2023-04-18 | 同济大学 | Imaging millimeter wave radar point cloud target classification method based on machine learning |
| CN113850308A (en) * | 2021-09-15 | 2021-12-28 | 温州大学大数据与信息技术研究院 | Target classification method for complex scene |
| CN114818916B (en) * | 2022-04-25 | 2023-04-07 | 电子科技大学 | Road target classification method based on millimeter wave radar multi-frame point cloud sequence |
| CN114972788A (en) * | 2022-05-25 | 2022-08-30 | 江汉大学 | Outlier extraction method and device of three-dimensional point cloud |
-
2022
- 2022-12-13 CN CN202211594912.7A patent/CN115586506B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN115586506A (en) | 2023-01-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11402494B2 (en) | Method and apparatus for end-to-end SAR image recognition, and storage medium | |
| Wheeler et al. | Deep stochastic radar models | |
| CN109087510B (en) | Traffic monitoring method and device | |
| US9286524B1 (en) | Multi-task deep convolutional neural networks for efficient and robust traffic lane detection | |
| US20230099113A1 (en) | Training method and apparatus for a target detection model, target detection method and apparatus, and medium | |
| DE102017119538A1 (en) | Physical modeling for radar and ultrasonic sensors | |
| Sameen et al. | A two-stage optimization strategy for fuzzy object-based analysis using airborne LiDAR and high-resolution orthophotos for urban road extraction | |
| CN112824931A (en) | Method and apparatus for improving radar data using reference data | |
| CN114022830A (en) | Target determination method and target determination device | |
| CN115586506B (en) | Anti-interference target classification method and device | |
| CN116027324B (en) | Fall detection method and device based on millimeter wave radar and millimeter wave radar equipment | |
| CN116468950A (en) | Three-dimensional target detection method for neighborhood search radius of class guide center point | |
| CN115965847A (en) | Three-dimensional target detection method and system based on multi-modal feature fusion under cross view angle | |
| CN112651405B (en) | Target detection method and device | |
| Rajender et al. | Application of Synthetic Aperture Radar (SAR) based Control Algorithms for the Autonomous Vehicles Simulation Environment | |
| CN116964472A (en) | Method for detecting at least one object of an environment by means of a reflected signal of a radar sensor system | |
| CN114882458A (en) | Target tracking method, system, medium and device | |
| CN115578608B (en) | Anti-interference classification method and device based on millimeter wave radar point cloud | |
| Prasvita et al. | Automatic detection of oil palm growth rate status with YOLOv5 | |
| Wachtel et al. | Convolutional neural network classification of vulnerable road users based on micro-doppler signatures using an automotive radar | |
| Prasanna | Sensor Fusion in Neural Networks for Object Detection | |
| Subash | Automatic road extraction from satellite images using extended Kalman filtering and efficient particle filtering | |
| US20240135195A1 (en) | Efficient search for data augmentation policies | |
| KR102659340B1 (en) | System and method for detecting information of a moving object based on a single Synthetic Aperture Radar (SAR) | |
| Rangkuti et al. | Development of Vehicle Detection and Counting Systems with UAV Cameras: Deep Learning and Darknet Algorithms |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |